Doctor inquiry auxiliary method and system based on large language model

Abstract

The invention provides a doctor inquiry assisting method and system based on a large language model, and belongs to the technical field of intelligent medical treatment. According to the scheme, the method comprises the following steps: collecting communication information between a doctor and a patient in an inquiry process, inputting the communication information into a trained large voice model, and obtaining keywords related to the condition of the patient in the communication information; obtaining a diagnosis result of the disease of the patient by using the keyword for reference of a doctor; generating a prescription suggestion for the doctor according to the diagnosis result; according to communication information and prescription data between doctors and patients in the inquiry process, medical records are preliminarily filled in, and the doctors finally confirm the medical records. Through the functions of intelligent guide type inquiry, historical case matching, diagnosis analysis, doctor assistance and the like, comprehensive inquiry assistance and diagnosis support are provided for doctors. The efficiency and accuracy of doctor inquiry can be remarkably improved, and more comprehensive and personalized medical services are provided for patients.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent medical technology, and in particular relates to a doctor consultation assistance method and system based on a large language model. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Inquiry about the condition and writing of medical records account for a large proportion of the doctor's consultation process. During the consultation process, doctors often need to spend a lot of working time browsing the patient's medical records, looking up the standard diagnosis and treatment process and medical guidelines for related diseases, and the indications and contraindications of drugs, so as to make a diagnosis and treatment plan that meets the patient's condition and diagnosis and treatment standards. The tedious query and reading work greatly affects the doctor's work efficiency and increases the doctor's workload. Especially in some tertiary hospitals, senior doctors have heavy outpatient tasks and the consultation time for a single patient is extremely limited. Young doctors and interns are often required to assist in the diagnosis and treatment, transforming the patient's long-term and complex description of the condition into a more valuable and standardized description of the condition from the doctor's perspective, and completing the writing of outpatient medical records. The repetitive and tedious assistant work also increases the burden on the assistant doctors. On the other hand, the diagnosis and treatment process and medical guidelines of clinical diseases are constantly updated, and different doctors have different understandings of the latest medical guidelines. Therefore, through the most cutting-edge computer technology, the automatic collection of diagnosis and treatment information, standard diagnosis and treatment procedures and medical guidelines for related diseases, and automatic matching and query reference of drug indications and contraindications can be realized, which will greatly improve the efficiency of doctors' diagnosis and treatment, greatly improve the accuracy and safety of doctors' medical decision-making, and save the human resources of assistant doctors.

[0004] In recent years, artificial intelligence technology, especially large language model technology, has flourished, making it possible for computers to intelligently analyze conversations and provide context-related auxiliary information. In the field of health care, large language models can automatically generate all clinical texts required for the entire diagnosis and treatment process, and assist clinical decision-making based on patients / problems, interventions, comparative measures, and outcomes in evidence-based medicine. Large language models (LLMs) can assist in diagnosis and treatment activities, providing doctors with instant and accurate information support in the process of diagnosis, treatment planning, patient communication, and case records, ultimately achieving the goal of improving the efficiency and quality of medical services. Summary of the invention

[0005] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a method and system for assisting doctors in consultation based on a large language model, which provides comprehensive consultation assistance and diagnosis support for doctors through intelligent guided consultation, historical case matching, diagnosis analysis and doctor assistance, etc. It can significantly improve the efficiency and accuracy of doctors' consultation and provide patients with more comprehensive and personalized medical services.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: The first aspect of the present invention provides a doctor consultation assistance method based on a large language model; A doctor consultation assistance method based on a large language model, comprising: Collecting the voice communication information between the doctor and the patient during the consultation process, inputting the voice communication information into the trained large language model, and obtaining keywords related to the patient's condition in the communication information; Obtain the diagnosis result of the patient's disease according to the keywords for the doctor's reference; Generate prescription suggestions for doctors based on the diagnosis results, the prescription suggestions including recommended test orders, drug selection, usage, and number of days; The medical record is filled in through the communication information between the doctor and the patient during the consultation process and the prescription data.

[0007] As a further technical solution, the communication information between the doctor and the patient during the consultation process is collected, and the communication information is input into the trained large language model. The process of obtaining keywords related to the patient's condition in the communication information is as follows: Collect the original audio data of doctors and patients during the consultation process, perform real-time voice transcription, and convert the original audio data into text data; Calculating the correlation between characters in the text data and other characters to obtain a boundary representation; Obtaining the dependency relationship between characters in the keywords in the text data to obtain an internal information representation; Performing feature fusion on the boundary representation and the internal information representation to obtain the predicted keyword type; According to the predicted keyword type, a predicted keyword type score is calculated; and keywords in the input text data are identified according to the predicted keyword type score.

[0008] As a further technical solution, the process of obtaining the diagnosis result of the patient's disease based on the keywords for the doctor's reference is: by calling the large language model that has been fine-tuned by medical standard documents and clinical diagnosis and treatment guidelines, the diagnosis result of the patient's disease is obtained according to the keywords; the diagnosis result includes: the name of the relevant disease, the required diagnosis and treatment process, necessary examinations, therapeutic drugs, and the conventional usage and dosage of the therapeutic drugs, indications, contraindications, common adverse reactions, and medical insurance reimbursement ratio prompts.

[0009] As a further technical solution, the process of generating prescription suggestions for doctors based on the diagnosis results is: automatically filling in the patient's basic information and diagnosis results into the prescription template; generating prescription suggestions for doctors through a comprehensive understanding of consultation records, medication guidelines, medical standard documents, medical records data and hospital pharmacy data through a large language model; after the doctor checks and improves the prescription, the prescription is uploaded to the prescription database.

[0010] As a further technical solution, the process of preliminarily filling in the medical record through the communication information between the doctor and the patient and the prescription data during the consultation process is: filling in the patient's basic information into the medical record template; summarizing the collected consultation records, and filling in the patient's key medical record information into the medical record template; integrating the patient's latest treatment plan and automatically filling in the medical record template; after the doctor checks and modifies the medical record, the medical record is updated in the medical record database.

[0011] As a further technical solution, low-rank adaptation LoRA is used to fine-tune the large language model, and a low-rank matrix is ​​introduced to update the weights of the pre-trained model, so that only a small number of parameters are modified during the fine-tuning process.

[0012] As a further technical solution, in the process of fine-tuning the large language model using low-rank adaptive LoRA, the effective weight W' of the original weight W in the large language model after LoRA adjustment is expressed as:

[0013] Among them, W is the weight of the original linear layer; A and B are newly added small matrices, which together constitute the low-rank update part; the product AB is a matrix of the same size as W but with rank r, which serves as an increment to W.

[0014] A second aspect of the present invention provides a doctor consultation assistance system based on a large language model.

[0015] A doctor consultation assistance system based on a large language model, comprising: The consultation voice information collection module is configured to: collect the voice communication information between the doctor and the patient during the consultation process, input the voice communication information into the trained large language model, and obtain keywords related to the patient's condition in the communication information; The diagnosis result output module is configured to: obtain the diagnosis result of the patient's disease according to the keyword for the doctor's reference; A prescription auxiliary module is configured to: generate prescription suggestions for doctors based on the diagnosis results, wherein the prescription suggestions include recommended test orders, drug selection, usage, and number of days; The medical record generation auxiliary module is configured to: preliminarily fill in the medical record through the communication information between the doctor and the patient and the prescription data during the consultation process, and the doctor will make the final confirmation of the medical record.

[0016] The third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of a doctor consultation assistance method based on a large language model as described in the first aspect of the present invention.

[0017] The fourth aspect of the present invention provides an electronic device, comprising a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in a doctor consultation assistance method based on a large language model as described in the first aspect of the present invention are implemented.

[0018] One or more of the above technical solutions have the following beneficial effects: The present invention provides doctors with comprehensive consultation assistance and diagnostic support through intelligent guided consultation, historical case matching, diagnostic analysis and doctor assistance. It can significantly improve the efficiency and accuracy of doctors' consultations and provide patients with more comprehensive and personalized medical services. At the same time, the system also has the ability of continuous optimization and training, which can continuously improve performance and reliability to meet the actual needs of the medical industry.

[0019] By using large language models (LLMs) to assist in diagnosis and treatment activities, doctors can be provided with immediate and accurate information support in the process of diagnosis, treatment planning, patient communication and case record keeping, ultimately achieving the goal of improving the efficiency and quality of medical services.

[0020] By fine-tuning the large language model through LoRA, while keeping most of the pre-trained weights unchanged, only relatively few parameters are added to adapt to new tasks, thereby reducing computational complexity and memory requirements, significantly reducing computational and storage costs while ensuring model performance.

[0021] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0023] Figure 1 This is a flow chart of the method of the first embodiment.

[0024] Figure 2 It is a system structure diagram of the second embodiment. DETAILED DESCRIPTION

[0025] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0026] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0027] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0028] The present invention provides doctors with comprehensive consultation assistance and diagnostic support through intelligent guided consultation, historical case matching, diagnostic analysis and doctor assistance. It can significantly improve the efficiency and accuracy of doctors' consultations and provide patients with more comprehensive and personalized medical services. The present application scheme will be further described below in conjunction with specific embodiments.

[0029] Embodiment 1 This embodiment discloses a doctor consultation assistance method based on a large language model; like Figure 1 As shown, a doctor consultation assistance method based on a large language model includes: Step S1, collecting voice communication information between the doctor and the patient during the consultation process, inputting the voice communication information into a trained large language model, and obtaining keywords related to the patient's condition in the communication information; In step S1, the communication information between the doctor and the patient during the consultation is collected, and the communication information is identified to obtain keywords related to the patient's condition. The communication information includes the doctor's inquiry of the patient's main complaint, current medical history and other medical information, that is, the symptoms that the patient feels most uncomfortable or most concerned about and their duration, as well as the patient's medical history, surgical and trauma history, allergy history, marital history, family history, as well as living habits, diet, etc.

[0030] In the process of obtaining keywords related to the patient's condition, it is necessary to correctly identify keyword boundaries and categories, where keyword boundaries are the starting and ending positions of keywords in text data, and keyword categories are the keyword categories to which keywords belong. When determining the boundaries, the model needs to take into account the contextual information of the keywords, and after the model determines the keyword boundaries, it is necessary to further extract the internal information of the keywords to enhance the recognition of the keyword categories. Therefore, step S1 also includes: Step S1.1, converting the voice communication information between the doctor and the patient into text data by real-time voice transcription.

[0031] Step S1.2, calculate the correlation between the character in the text data and other characters to obtain the boundary representation; wherein the character weight is calculated using the self-attention mechanism to calculate the correlation between the current character and other characters. As input, calculate the weight of each character: ; Where Q, K are the query matrices in the attention mechanism, V is the key-value matrix, is the number of dimensions of h.

[0032] Multiple self-attention mechanisms are concatenated to obtain an enhanced representation of word vectors. Based on the word vector sequence after enhancing the boundary information, the Biaffine attention mechanism is used to obtain the boundary representation.

[0033] ; in is the vector representation of the span head and tail, Represents a splicing operation, are two trainable parameters, b is the bias, is the boundary information representation of ij, and T is the transposed symbol.

[0034] Step S1.3, obtain the dependency relationship between words in the keywords in the text data to obtain the internal information representation; extract the dependency information between words in the keywords through conditional layer normalization (CLN). Conditional layer normalization is a deep learning technology. On the basis of layer normalization, the normalization parameters are dynamically adjusted according to the data information input by a specific task, which enables the model to be adaptively adjusted for different data distributions. This can solve the gradient vanishing and gradient exploding problems in deep neural network training. It accelerates the convergence speed of the model by standardizing the input of each layer of the neural network and can improve the network's expression ability. The formula is as follows: ; ; ; In the formula, , is the word vector at position i, j in the span, is the influence of the character at index position i on the character at index position j, , is a trainable parameter, , is the bias, and are the mean and standard deviation of all character weights, respectively, and the calculation formula is: ; ; In the formula, is the number of dimensions of h, represent The kth dimension of .

[0035] After obtaining the dependency information between all words in the keyword, in order to fuse all the dependency information, the interaction information obtained at each index position is added to the word vector to obtain a new word vector sequence. In order to highlight the importance of different characters in the keyword, a linear attention mechanism is used to fuse all word vectors in the span. The formula is: ； ; ; ; In the formula, is a word vector that integrates the relationship between words. is the span length, t is the index of the character vector, It represents the internal information of the span.

[0036] Step S1.4, feature fusion of the boundary representation and the internal information representation to obtain the predicted keyword type; wherein a gating mechanism is used to fuse the boundary representation and the internal information representation, wherein the gating mechanism formula is: ; Where U is the time gate, b is the bias, It is a sigmoid function; Use a linear classifier and SoftMax function to predict the entity type of each span. The formula is: ; ; In the formula, is the span representation, is a trainable parameter, is bias.

[0037] Step S1.5, according to the predicted keyword type, calculate the predicted keyword type score; identify the keywords in the input text data according to the predicted keyword type score. The LOSS function is also introduced to score the keywords, where the Loss function is a cross entropy loss function: ; In the formula, M is the number of training samples, is the prediction type value, is the true type value. During training, the prediction output selects the keyword with the highest score and takes the keyword with the highest score as the recognized keyword.

[0038] Step S2, input the keywords into the trained large language model again to obtain the diagnosis results of the patient's disease for the doctor's reference; the diagnosis results include: the name of the relevant disease, the required diagnosis and treatment process, necessary examinations, therapeutic drugs, and the conventional usage and dosage of the therapeutic drugs, indications, contraindications, common adverse reactions, and medical insurance reimbursement ratio prompts.

[0039] In step S1 and step S2, the large language models are trained with large-scale data sets and are able to understand and generate various types of texts, such as articles, conversations, poems, codes, etc. By learning a large amount of language patterns and contextual information, they are able to complete a variety of natural language processing tasks, such as answering questions, translating, summarizing text, writing stories, providing suggestions, and generating code.

[0040] In this embodiment, the large language model is Qwen2-7B-Instruct, which is a medium-sized language model created by Alibaba Cloud. It has been specially fine-tuned to perform instruction-based tasks well. It has 2.7 billion parameters. The uniqueness of this model is that it has undergone a deep fine-tuning phase, especially in instruction-oriented tasks, which enables it to accurately capture and respond to users' specific needs. On the technical level, the model relies on the Transformer architecture. This revolutionary technology ensures that the model can efficiently handle long-distance dependencies and understand the nuances in the context, thereby producing coherent, in-depth responses that are close to human thinking logic. In addition, through continuous learning and optimization, Qwen2-7B-Instruct is continuously improving the diversity and innovation of its generated content while protecting user privacy and ensuring content security.

[0041] In this embodiment, the large language model is also fine-tuned. Specifically, the fine-tuning data mainly consists of medical standard documents, clinical diagnosis and treatment guidelines, and hospital pharmacy data.

[0042] For medical standard documents, people with professional medical knowledge manually annotate the documents, and the keyword types include: disease_name, treatment_ methods, treatment_plan, and applicable_population. After the annotation is completed, the tag information in the document is processed into JSON format, and the file name is data1.json. Disease_name is a composite key, including secondary keys treatment_ methods, treatment_plan, and applicable_population.

[0043] For the treatment of clinical medication guidelines, the therapeutic drug types include: drug_name, drug_classification, contraindications, adverse_reactions, usage, and dosage.

[0044] After the annotation is completed, the tag information in the document is processed into JSON format, the file name is data2.json, drug_name is a composite key, including secondary keys drug_ classification, contraindications, adverse_reactions, usage, dosage. For the drug treatment principles in the document, the types are divided into disease_name, drug_name, and precautions. Processed into JSON format, the file name is data3.json, disease_name is a composite key, including secondary keys drug_name and precautions.

[0045] First, the raw data must be labeled and processed to meet the data quality and format requirements of the large language model fine-tuning function. The fine-tuning data processing methods and strategies mainly include data cleaning and preprocessing, data enhancement, and data partitioning.

[0046] For different large language model products, the fine-tuning methods are also different, such as reading fine-tuning data files locally, or calling fine-tuning APIs remotely, etc. When choosing a fine-tuning solution, considering parameter efficiency, flexibility, ease of use, generalization ability, scalability, and multi-tasking, Low-Rank Adaptation LoRA (Low-Rank Adaptation) has some unique advantages over other PEFT methods.

[0047] LoRA is a parameter-efficient fine-tuning method whose core idea is to update the weights of the pre-trained model by introducing a low-rank matrix, so that only a small number of parameters are modified during the fine-tuning process. LoRA allows the model to adapt to new tasks by only adding relatively few parameters while keeping most of the pre-trained weights unchanged, thereby reducing computational complexity and memory requirements.

[0048] The original weight W of the large language model is adjusted by LoRA, and the effective weight W' can be expressed as:

[0049] Among them, W is the weight of the original linear layer; A and B are newly added small matrices, which together constitute the low-rank update part; the product AB is a matrix of the same size as W but with rank r, which serves as an increment to W.

[0050] The process of fine-tuning a large language model using low-rank adaptation LoRA is: For the input fine-tuning data, Qwen2-7B-Instruct generates a word vector sequence , and classify the words into M, B, S, and E sets according to the character position, where the M set represents the words that store characters in the middle position, the B set represents the words that store characters in the first position, the S set represents the words that store characters, and the E set represents the words that store characters in the tail position.

[0051] The weight of each word in the word set is calculated through the linear attention mechanism, and the vocabulary information in the weighted set word set is calculated using the formula:

[0052]

[0053]

[0054] in , is the vocabulary index in the word set. After dynamically weighting the word set representation, the four word sets are concatenated to represent the final representation of each character:

[0055] Pass the embedded input data to the model for forward propagation. In LoRA fine-tuning, the model uses the embedded input data and the parameters of the LoRA adapter for calculation. The forward propagation process of the model is updated as:

[0056] Since the keys in each JSON file are different, fine-tuning with the same parameters will cause the accuracy of the large language model to decrease. Therefore, the three documents are fine-tuned separately, and the fine-tuned qwen2-7B-instruct model achieves the best performance by changing the LoRA configuration. The training parameters of each round remain unchanged, and the fine-tuning order is data1.json, data2.json, and data3.json.

[0057] The first round of fine-tuning uses a smaller `r` value and a higher `lora_dropout`, which can prevent premature overfitting while keeping the model simple and quickly obtain a basic model; the second round of fine-tuning increases the `r` value and reduces `lora_dropout`, allowing the model to learn more details while still retaining a certain regularization; the third round of fine-tuning further increases the `r` value and reduces `lora_dropout`. The model has obtained enough information from the previous rounds and can focus on fine-tuning to achieve optimal performance.

[0058] When fine-tuning data1.json, LoRA configuration r=4, lora_alpha=16, target_modules=["query","key","value"], lora_dropout=0.1.

[0059] When fine-tuning data2.json, LoRA configuration r=8, lora_alpha=32, target_modules=["query","key","value"], lora_dropout=0.05.

[0060] When fine-tuning data3.json, LoRA configuration r=16, lora_alpha=64, target_modules=["query","key","value"], lora_dropout=0.01.

[0061] Where r is the rank of the LoRA matrix decomposition, lora_alpha is used to scale the LoRA weights, lora_dropout is used for the dropout ratio of the LoRA layer, which helps prevent overfitting, and target_modules specifies which modules or layers will be applied to the LoRA adapter.

[0062] After training is complete, evaluate the performance of the model and save the fine-tuned model.

[0063] Step S3, based on the diagnosis results, a prescription suggestion is generated for the doctor, and the prescription suggestion includes the recommended test order, drug selection, usage, and number of days. In step S3, the patient's basic information is automatically filled in the prescription; the doctor searches for medical information, and the prescription assistance interface provides relevant drug options and displays drug information (such as name, dosage, usage, brand, etc.). The doctor selects the examination and drug through decision-making, as well as the dosage, usage, and intended number of days of the drug; the data collected through the dialogue information is processed and filled in the prescription; the doctor clicks the preview button to view the complete prescription. After confirming that it is correct, the doctor clicks the submit button, and the prescription will be uploaded to the database and synchronized to the relevant examination department, pharmacy, etc.

[0064] Step S4, the medical record is preliminarily filled out based on the communication information between the doctor and the patient and the prescription data during the consultation process, and the doctor makes the final confirmation of the medical record.

[0065] In step S4, the basic information of the patient is filled into the medical record template; the collected consultation records are summarized and the key information of the patient's medical record is filled into the medical record template; the patient's latest treatment plan is integrated and automatically filled into the medical record template; after the doctor checks and modifies the medical record, the medical record is updated in the medical record database. Embodiment 2 This embodiment discloses a doctor consultation assistance system based on a large language model; like Figure 2 As shown, a doctor consultation assistance system based on a large language model includes: The consultation voice information collection module is configured to: collect the voice communication information between the doctor and the patient during the consultation process, input the voice communication information into the trained large language model, and obtain keywords related to the patient's condition in the communication information; The diagnosis result output module is configured to: obtain the diagnosis result of the patient's disease according to the keyword for the doctor's reference; A prescription auxiliary module is configured to: generate prescription suggestions for doctors based on the diagnosis results, wherein the prescription suggestions include recommended test orders, drug selection, usage, and number of days; The medical record generation auxiliary module is configured to: preliminarily fill in the medical record through the communication information between the doctor and the patient and the prescription data during the consultation process, and the doctor will make the final confirmation of the medical record.

[0066] Embodiment 3 The purpose of this embodiment is to provide a computer-readable storage medium.

[0067] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a doctor consultation assistance method based on a large language model as described in Example 1 of the present disclosure.

[0068] Embodiment 4 The purpose of this embodiment is to provide an electronic device.

[0069] An electronic device comprises a memory, a processor and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in a method for assisting a doctor's consultation based on a large language model as described in Example 1 of the present disclosure are implemented.

[0070] The steps involved in the apparatuses of the above embodiments 2, 3 and 4 correspond to the method embodiment 1, and the specific implementation methods can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0071] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0072] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A doctor consultation assistance method based on a large language model, characterized in that: include: Collecting the voice communication information between the doctor and the patient during the consultation process, inputting the voice communication information into the trained large language model, and obtaining keywords related to the patient's condition in the communication information; Obtain the diagnosis result of the patient's disease according to the keywords for the doctor's reference; Generate prescription suggestions for doctors based on the diagnosis results, the prescription suggestions including recommended test orders, drug selection, usage, and number of days; The medical record is filled in through the communication information between the doctor and the patient during the consultation process and the prescription data.

2. A doctor consultation assistance method based on a large language model as claimed in claim 1, characterized in that: The process of collecting the communication information between the doctor and the patient during the consultation process, inputting the communication information into the trained large language model, and obtaining the keywords related to the patient's condition in the communication information is as follows: Collect the original audio data of doctors and patients during the consultation process, perform real-time voice transcription, and convert the original audio data into text data; Calculating the correlation between characters in the text data and other characters to obtain a boundary representation; Obtaining the dependency relationship between characters in the keywords in the text data to obtain an internal information representation; Performing feature fusion on the boundary representation and the internal information representation to obtain the predicted keyword type; According to the predicted keyword type, a predicted keyword type score is calculated; and keywords in the input text data are identified according to the predicted keyword type score.

3. A doctor consultation assistance method based on a large language model as claimed in claim 1, characterized in that: The process of obtaining the diagnosis result of the patient's disease for the doctor's reference according to the keywords is as follows: obtaining the diagnosis result of the patient's disease according to the keywords by calling the large language model fine-tuned by the medical standard documents and clinical diagnosis and treatment guidelines; The diagnosis results include: the name of the relevant disease, the required diagnosis and treatment process, necessary examinations, therapeutic drugs, and the conventional usage and dosage of the therapeutic drugs, indications, contraindications, common adverse reactions, and medical insurance reimbursement ratio.

4. A doctor consultation assistance method based on a large language model as claimed in claim 1, characterized in that: The process of generating prescription suggestions for doctors based on the diagnosis results is as follows: automatically filling in the patient's basic information and diagnosis results into the prescription template; generating prescription suggestions for doctors through a comprehensive understanding of consultation records, medication guidelines, medical standard documents, medical record data and hospital pharmacy data through a large language model; and uploading the prescription to the prescription database after the doctor checks and improves the prescription.

5. The method for assisting doctor consultation based on a large language model as claimed in claim 1, characterized in that: The process of initially filling out the medical record through the communication information between the doctor and the patient and the prescription data during the consultation process is as follows: filling in the patient's basic information into the medical record template; summarizing the collected consultation records and filling in the patient's key medical record information into the medical record template; integrating the patient's latest treatment plan and automatically filling in the medical record template; after the doctor checks and modifies the medical record, the medical record is updated in the medical record database.

6. A doctor consultation assistance method based on a large language model as claimed in claim 1, characterized in that: Low-rank adaptation LoRA is used to fine-tune the large language model, and a low-rank matrix is ​​introduced to update the weights of the pre-trained model, so that only a small number of parameters are modified during the fine-tuning process.

7. A doctor consultation assistance method based on a large language model as claimed in claim 6, characterized in that: In the process of fine-tuning the large language model using low-rank adaptation LoRA, the effective weight W' after the original weight W in the large language model is adjusted by LoRA is expressed as: Among them, W is the weight of the original linear layer; A and B are newly added small matrices, which together constitute the low-rank update part; the product AB is a matrix of the same size as W but with rank r, which serves as an increment to W.

8. A doctor consultation assistance system based on a large language model, characterized in that: include: The consultation voice information collection module is configured to: collect the voice communication information between the doctor and the patient during the consultation process, input the voice communication information into the trained large language model, and obtain keywords related to the patient's condition in the communication information; The diagnosis result output module is configured to: obtain the diagnosis result of the patient's disease according to the keyword for the doctor's reference; A prescription auxiliary module is configured to: generate prescription suggestions for doctors based on the diagnosis results, wherein the prescription suggestions include recommended test orders, drug selection, usage, and number of days; The medical record generation auxiliary module is configured to: preliminarily fill in the medical record through the communication information between the doctor and the patient and the prescription data during the consultation process, and the doctor will make the final confirmation of the medical record.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps in the doctor consultation assistance method based on a large language model as described in any one of claims 1 to 7 are implemented.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the doctor consultation assistance method based on a large language model as described in any one of claims 1 to 7 are implemented.

Images